arc-reactor

Arc-Reactor

AI Company OS — one core system that powers all agent teams simultaneously.

Arc-Reactor is a multi-team AI orchestration engine that decomposes high-level goals into team-specific tasks, executes them in parallel, and validates results through quality gates.

Inspired by Iron Man's Arc Reactor — a single power source that runs everything.

How It Works

You: "Build a login page with email/password auth"
         │
         ▼
    Director (Opus)
    ├── Analyzes goal
    ├── Identifies features (planning → dev-level)
    └── Creates execution plan with waves
         │
         ▼
    Architect Agent
    └── Reviews plan for cross-cutting concerns
         │
         ▼
    Wave 1 (parallel)
    ├── [Frontend] Login form component
    ├── [Backend]  Auth API endpoint
    └── [Design]   Login page design tokens
         │
         ▼
    Wave 2 (parallel)
    ├── [QA]       E2E tests for login flow
    ├── [Security] Auth vulnerability review
    └── [Docs]     API documentation
         │
         ▼
    Quality Gate
    ├── Code generated?  ✅
    ├── Tests passing?   ✅
    └── No conflicts?    ✅
         │
         ▼
    Git Workflow (optional)
    ├── Create feature branch
    ├── Auto-commit results
    └── Push & create PR
         │
         ▼
    ✅ Done — files created, tests passing

Quick Start

# From your project directory
cd my-project

# Run with Claude Code subscription (default — subagent mode)
node /path/to/arc-reactor/packages/cli/dist/index.js ignite "Build a todo app"

# Or with API key
ANTHROPIC_API_KEY=sk-... node /path/to/arc-reactor/packages/cli/dist/index.js ignite "Build a todo app" --mode api

# With git workflow
arc-reactor ignite "Build user profile" \
  --feature-id user-profile \
  --auto-branch \
  --auto-commit \
  --create-pr

CLI Commands

arc-reactor ignite <goal>

Start orchestration for a goal.

arc-reactor ignite "Build a login page with email/password auth"
arc-reactor ignite "Create a REST API for user management" --mode api
arc-reactor ignite "Build a dashboard" --teams frontend,backend,security
arc-reactor ignite "Add OAuth2" --feature-id oauth2 --auto-branch --auto-commit --create-pr

Options:

Flag	Description	Default
--mode <mode>	subagent (Claude CLI), api (API key), auto	subagent
--teams <list>	Comma-separated team list	frontend,backend,qa
--verbose	Enable verbose output	false
--auto-branch	Create a feature branch (feature/{id})	false
--auto-commit	Auto-commit generated files after success	false
--create-pr	Push branch and create PR after commit	false
--feature-id <id>	Feature ID for branch naming and Vibranium tracking	—

arc-reactor config

Show or edit configuration.

arc-reactor config                          # Show current config
arc-reactor config --set model=claude-sonnet-4-6  # Set a value

Architecture

arc-reactor/
├── packages/
│   ├── core/               # Standalone orchestration engine
│   │   ├── orchestrator/   # Director, Architect, task decomposition, wave scheduling
│   │   ├── teams/          # 8 team definitions (Frontend, Backend, QA, Design, DevOps, Security, Docs, Product)
│   │   ├── executor/       # Dual executor: subagent (default) + API
│   │   ├── quality-gate/   # Code check, test runner, conflict detection
│   │   └── git-ops.ts      # Git workflow (branch, commit, push, PR)
│   └── cli/                # CLI wrapper (commander)
├── plugins/arc-reactor/    # Claude Code plugin (marketplace structure)
│   ├── agents/             # 10 agent definitions (Director, Architect, 8 teams)
│   └── hooks/              # Vibranium hook integration (6 hooks)
└── .mcp.json               # Vibranium MCP server config

Execution Modes

Mode	How	Cost	Requires
subagent (default)	Spawns claude CLI per task	Subscription included	Claude Code installed
api	Direct Anthropic SDK calls	API billing	ANTHROPIC_API_KEY
auto	Prefers subagent, falls back to API	—	Either

Teams (v0.5)

Team	Specialization
Frontend	React/Next.js components, responsive UI, accessibility
Backend	REST APIs, auth, database, business logic
QA	Unit tests, E2E tests, edge cases
Design	Design tokens, component styling, responsive layouts
DevOps	CI/CD pipelines, Docker, deployment configs
Security	Vulnerability assessment, OWASP review, auth audit
Docs	API documentation, guides, architecture docs
Product	Requirements analysis, user stories, feature specs

Architect Agent

The Architect agent reviews the Director's execution plan before task dispatch:

• Cross-cutting concern identification
• Architecture consistency checks
• Dependency validation